Troubleshooting Guide
Last updated July 1, 2025
What does "you have reached context limit" error mean?
The "context limit" error indicates your project has grown too large to fit within a single agent context. This typically happens with complex applications or extensive conversation history, as the limit exists to ensure the agent maintains a complete understanding of your project. To resolve this, please follow our detailed guidelines which includes techniques for breaking larger projects into manageable modules and properly saving progress to continue development in new tasks.
To find how to save your progress to GitHub and start a new chat, please follow the steps listed in the Features and Tools article at < https://atlas-kb.com/atlas-e74243keac/articles/549082-features-and-tools >
Agent is in sleep mode
When your Emergent AI agent enters sleep mode to conserve resources:
- Locate and click the "Wake Up Agent" button in the interface to reactivate the agent
- Allow up to 15 minutes for the agent to initialize if it doesn't activate immediately - this is a normal part of our system's resource management
- If the agent remains unresponsive after 1-3 minutes, please contact our technical support team at support@emergent.sh with details of your project
Agent waiting for human response
When you see the "waiting for human response" message:
- This indicates the agent needs your input to proceed with development
- Wait approximately one minute as sometimes there might be a slight processing delay
- If the agent continues waiting, simply type "continue" in the chat to prompt the agent to resume its work, or give additional instructions if you like
- This typically happens when the agent needs confirmation or additional information to proceed with your coding project
Agent keeps looping in the same question/issue
If your Emergent agent gets stuck in a loop or becomes unresponsive:
- Immediate actions you can take: Use the pause button located on the chat screen to halt the agent's current processes. Type "finish" in the chat to instruct the agent to complete its current task. Hover over the chat interface and use the rollback feature to return to a previous, working state.
- For preserving your work: You can use the "save to GitHub" function within the chat window to ensure your code is safely stored. This allows you to restart the agent without losing progress on your development
These issues occasionally occur when the agent encounters complex logic or ambiguous instructions. After resolving, try providing more specific guidance to avoid similar loops.
Code generation is very slow
Emergent's code generation process is deliberately thorough and may seem slower than simpler code generators because:
- Advanced development approach:
- Our AI-powered platform performs deep research and planning for each development task The agent creates complete, working applications with properly integrated frontend and backend components This comprehensive approach requires more processing time but delivers higher quality results
- What's happening behind the scenes:
- The agent is analyzing requirements, planning architecture, and considering best practices It's writing code, testing for errors, and optimizing as it progresses For complex applications, this process can take time but yields production-ready code
While we're continuously improving performance in our beta phase, we prioritize delivering functional applications rather than just quick code snippets. Your patience results in more robust, complete solutions.
How do I use rollback?
To use the rollback feature effectively:
- Click the rollback button in the chat interface
- Select the message to roll back to
- Choose either:
- "Erase all messages and generated code" (complete reset) "Erase messages only" (keeps code but removes conversation)
- Confirm your choice (note: this action cannot be undone)
Why am I being charged when I am not getting the output I want?
We understand why it feels unfair to spend credits when the agent’s first attempt isn’t perfect. In practice, though, each request to Emergent triggers real compute—parsing your prompt, writing code, compiling, running tests, and so on. That work consumes resources whether the outcome is exactly what you hoped for or needs refinement, just as an engineer’s billable hours cover both writing and debugging their code.
The good news is that “mistakes” are usually solvable iterations, not wasted effort. A few tips can minimize retries and credit use:
- Break large builds into smaller steps. Complete the backend first, then UI, then integrations—each stage is easier to debug and cheaper to rerun.
- Export checkpoints to GitHub often. This preserves working versions so you never have to rebuild from scratch.
- Follow our Best-Practices Guide for prompt structure, context limits, and common troubleshooting techniques: https://atlas-kb.com/atlas-e74243keac/articles/415975-best-practices
If the agent behaves unpredictably even after these steps, please share the run-ID with us. We’ll investigate, suggest a fix path, and—if a platform defect is confirmed—make things right.
Our goal is for every credit you spend to move your project forward, and we’re here to help you get the results you need.
I encountered an error not covered in the help section
If you've encountered an issue not addressed in our documentation:
- First check all help documentation for potential solutions
If your issue isn't addressed, email support@emergent.sh with:
- Screenshot of the error
- Your job ID
- Steps to reproduce the issue
- Description of what you were trying to do
For any questions or assistance, feel free to email us at support@emergent.sh